compact-agent 1.10.1 → 1.11.1

package/resources/ecc/skills/content-hash-cache-pattern/SKILL.md

@@ -0,0 +1,161 @@
+---
+name: content-hash-cache-pattern
+description: Cache expensive file processing results using SHA-256 content hashes — path-independent, auto-invalidating, with service layer separation.
+origin: ECC
+---
+
+# Content-Hash File Cache Pattern
+
+Cache expensive file processing results (PDF parsing, text extraction, image analysis) using SHA-256 content hashes as cache keys. Unlike path-based caching, this approach survives file moves/renames and auto-invalidates when content changes.
+
+## When to Activate
+
+- Building file processing pipelines (PDF, images, text extraction)
+- Processing cost is high and same files are processed repeatedly
+- Need a `--cache/--no-cache` CLI option
+- Want to add caching to existing pure functions without modifying them
+
+## Core Pattern
+
+### 1. Content-Hash Based Cache Key
+
+Use file content (not path) as the cache key:
+
+```python
+import hashlib
+from pathlib import Path
+
+_HASH_CHUNK_SIZE = 65536  # 64KB chunks for large files
+
+def compute_file_hash(path: Path) -> str:
+    """SHA-256 of file contents (chunked for large files)."""
+    if not path.is_file():
+        raise FileNotFoundError(f"File not found: {path}")
+    sha256 = hashlib.sha256()
+    with open(path, "rb") as f:
+        while True:
+            chunk = f.read(_HASH_CHUNK_SIZE)
+            if not chunk:
+                break
+            sha256.update(chunk)
+    return sha256.hexdigest()
+```
+
+**Why content hash?** File rename/move = cache hit. Content change = automatic invalidation. No index file needed.
+
+### 2. Frozen Dataclass for Cache Entry
+
+```python
+from dataclasses import dataclass
+
+@dataclass(frozen=True, slots=True)
+class CacheEntry:
+    file_hash: str
+    source_path: str
+    document: ExtractedDocument  # The cached result
+```
+
+### 3. File-Based Cache Storage
+
+Each cache entry is stored as `{hash}.json` — O(1) lookup by hash, no index file required.
+
+```python
+import json
+from typing import Any
+
+def write_cache(cache_dir: Path, entry: CacheEntry) -> None:
+    cache_dir.mkdir(parents=True, exist_ok=True)
+    cache_file = cache_dir / f"{entry.file_hash}.json"
+    data = serialize_entry(entry)
+    cache_file.write_text(json.dumps(data, ensure_ascii=False), encoding="utf-8")
+
+def read_cache(cache_dir: Path, file_hash: str) -> CacheEntry | None:
+    cache_file = cache_dir / f"{file_hash}.json"
+    if not cache_file.is_file():
+        return None
+    try:
+        raw = cache_file.read_text(encoding="utf-8")
+        data = json.loads(raw)
+        return deserialize_entry(data)
+    except (json.JSONDecodeError, ValueError, KeyError):
+        return None  # Treat corruption as cache miss
+```
+
+### 4. Service Layer Wrapper (SRP)
+
+Keep the processing function pure. Add caching as a separate service layer.
+
+```python
+def extract_with_cache(
+    file_path: Path,
+    *,
+    cache_enabled: bool = True,
+    cache_dir: Path = Path(".cache"),
+) -> ExtractedDocument:
+    """Service layer: cache check -> extraction -> cache write."""
+    if not cache_enabled:
+        return extract_text(file_path)  # Pure function, no cache knowledge
+
+    file_hash = compute_file_hash(file_path)
+
+    # Check cache
+    cached = read_cache(cache_dir, file_hash)
+    if cached is not None:
+        logger.info("Cache hit: %s (hash=%s)", file_path.name, file_hash[:12])
+        return cached.document
+
+    # Cache miss -> extract -> store
+    logger.info("Cache miss: %s (hash=%s)", file_path.name, file_hash[:12])
+    doc = extract_text(file_path)
+    entry = CacheEntry(file_hash=file_hash, source_path=str(file_path), document=doc)
+    write_cache(cache_dir, entry)
+    return doc
+```
+
+## Key Design Decisions
+
+| Decision | Rationale |
+|----------|-----------|
+| SHA-256 content hash | Path-independent, auto-invalidates on content change |
+| `{hash}.json` file naming | O(1) lookup, no index file needed |
+| Service layer wrapper | SRP: extraction stays pure, cache is a separate concern |
+| Manual JSON serialization | Full control over frozen dataclass serialization |
+| Corruption returns `None` | Graceful degradation, re-processes on next run |
+| `cache_dir.mkdir(parents=True)` | Lazy directory creation on first write |
+
+## Best Practices
+
+- **Hash content, not paths** — paths change, content identity doesn't
+- **Chunk large files** when hashing — avoid loading entire files into memory
+- **Keep processing functions pure** — they should know nothing about caching
+- **Log cache hit/miss** with truncated hashes for debugging
+- **Handle corruption gracefully** — treat invalid cache entries as misses, never crash
+
+## Anti-Patterns to Avoid
+
+```python
+# BAD: Path-based caching (breaks on file move/rename)
+cache = {"/path/to/file.pdf": result}
+
+# BAD: Adding cache logic inside the processing function (SRP violation)
+def extract_text(path, *, cache_enabled=False, cache_dir=None):
+    if cache_enabled:  # Now this function has two responsibilities
+        ...
+
+# BAD: Using dataclasses.asdict() with nested frozen dataclasses
+# (can cause issues with complex nested types)
+data = dataclasses.asdict(entry)  # Use manual serialization instead
+```
+
+## When to Use
+
+- File processing pipelines (PDF parsing, OCR, text extraction, image analysis)
+- CLI tools that benefit from `--cache/--no-cache` options
+- Batch processing where the same files appear across runs
+- Adding caching to existing pure functions without modifying them
+
+## When NOT to Use
+
+- Data that must always be fresh (real-time feeds)
+- Cache entries that would be extremely large (consider streaming instead)
+- Results that depend on parameters beyond file content (e.g., different extraction configs)

package/resources/ecc/skills/context-budget/SKILL.md

@@ -0,0 +1,135 @@
+---
+name: context-budget
+description: Audits Claude Code context window consumption across agents, skills, MCP servers, and rules. Identifies bloat, redundant components, and produces prioritized token-savings recommendations.
+origin: ECC
+---
+
+# Context Budget
+
+Analyze token overhead across every loaded component in a Claude Code session and surface actionable optimizations to reclaim context space.
+
+## When to Use
+
+- Session performance feels sluggish or output quality is degrading
+- You've recently added many skills, agents, or MCP servers
+- You want to know how much context headroom you actually have
+- Planning to add more components and need to know if there's room
+- Running `/context-budget` command (this skill backs it)
+
+## How It Works
+
+### Phase 1: Inventory
+
+Scan all component directories and estimate token consumption:
+
+**Agents** (`agents/*.md`)
+- Count lines and tokens per file (words × 1.3)
+- Extract `description` frontmatter length
+- Flag: files >200 lines (heavy), description >30 words (bloated frontmatter)
+
+**Skills** (`skills/*/SKILL.md`)
+- Count tokens per SKILL.md
+- Flag: files >400 lines
+- Check for duplicate copies in `.agents/skills/` — skip identical copies to avoid double-counting
+
+**Rules** (`rules/**/*.md`)
+- Count tokens per file
+- Flag: files >100 lines
+- Detect content overlap between rule files in the same language module
+
+**MCP Servers** (`.mcp.json` or active MCP config)
+- Count configured servers and total tool count
+- Estimate schema overhead at ~500 tokens per tool
+- Flag: servers with >20 tools, servers that wrap simple CLI commands (`gh`, `git`, `npm`, `supabase`, `vercel`)
+
+**CLAUDE.md** (project + user-level)
+- Count tokens per file in the CLAUDE.md chain
+- Flag: combined total >300 lines
+
+### Phase 2: Classify
+
+Sort every component into a bucket:
+
+| Bucket | Criteria | Action |
+|--------|----------|--------|
+| **Always needed** | Referenced in CLAUDE.md, backs an active command, or matches current project type | Keep |
+| **Sometimes needed** | Domain-specific (e.g. language patterns), not referenced in CLAUDE.md | Consider on-demand activation |
+| **Rarely needed** | No command reference, overlapping content, or no obvious project match | Remove or lazy-load |
+
+### Phase 3: Detect Issues
+
+Identify the following problem patterns:
+
+- **Bloated agent descriptions** — description >30 words in frontmatter loads into every Task tool invocation
+- **Heavy agents** — files >200 lines inflate Task tool context on every spawn
+- **Redundant components** — skills that duplicate agent logic, rules that duplicate CLAUDE.md
+- **MCP over-subscription** — >10 servers, or servers wrapping CLI tools available for free
+- **CLAUDE.md bloat** — verbose explanations, outdated sections, instructions that should be rules
+
+### Phase 4: Report
+
+Produce the context budget report:
+
+```
+Context Budget Report
+═══════════════════════════════════════
+
+Total estimated overhead: ~XX,XXX tokens
+Context model: Claude Sonnet (200K window)
+Effective available context: ~XXX,XXX tokens (XX%)
+
+Component Breakdown:
+┌─────────────────┬────────┬───────────┐
+│ Component       │ Count  │ Tokens    │
+├─────────────────┼────────┼───────────┤
+│ Agents          │ N      │ ~X,XXX    │
+│ Skills          │ N      │ ~X,XXX    │
+│ Rules           │ N      │ ~X,XXX    │
+│ MCP tools       │ N      │ ~XX,XXX   │
+│ CLAUDE.md       │ N      │ ~X,XXX    │
+└─────────────────┴────────┴───────────┘
+
+WARNING: Issues Found (N):
+[ranked by token savings]
+
+Top 3 Optimizations:
+1. [action] → save ~X,XXX tokens
+2. [action] → save ~X,XXX tokens
+3. [action] → save ~X,XXX tokens
+
+Potential savings: ~XX,XXX tokens (XX% of current overhead)
+```
+
+In verbose mode, additionally output per-file token counts, line-by-line breakdown of the heaviest files, specific redundant lines between overlapping components, and MCP tool list with per-tool schema size estimates.
+
+## Examples
+
+**Basic audit**
+```
+User: /context-budget
+Skill: Scans setup → 16 agents (12,400 tokens), 28 skills (6,200), 87 MCP tools (43,500), 2 CLAUDE.md (1,200)
+       Flags: 3 heavy agents, 14 MCP servers (3 CLI-replaceable)
+       Top saving: remove 3 MCP servers → -27,500 tokens (47% overhead reduction)
+```
+
+**Verbose mode**
+```
+User: /context-budget --verbose
+Skill: Full report + per-file breakdown showing planner.md (213 lines, 1,840 tokens),
+       MCP tool list with per-tool sizes, duplicated rule lines side by side
+```
+
+**Pre-expansion check**
+```
+User: I want to add 5 more MCP servers, do I have room?
+Skill: Current overhead 33% → adding 5 servers (~50 tools) would add ~25,000 tokens → pushes to 45% overhead
+       Recommendation: remove 2 CLI-replaceable servers first to stay under 40%
+```
+
+## Best Practices
+
+- **Token estimation**: use `words × 1.3` for prose, `chars / 4` for code-heavy files
+- **MCP is the biggest lever**: each tool schema costs ~500 tokens; a 30-tool server costs more than all your skills combined
+- **Agent descriptions are loaded always**: even if the agent is never invoked, its description field is present in every Task tool context
+- **Verbose mode for debugging**: use when you need to pinpoint the exact files driving overhead, not for regular audits
+- **Audit after changes**: run after adding any agent, skill, or MCP server to catch creep early

package/resources/ecc/skills/continuous-agent-loop/SKILL.md

@@ -0,0 +1,45 @@
+---
+name: continuous-agent-loop
+description: Patterns for continuous autonomous agent loops with quality gates, evals, and recovery controls.
+origin: ECC
+---
+
+# Continuous Agent Loop
+
+This is the v1.8+ canonical loop skill name. It supersedes `autonomous-loops` while keeping compatibility for one release.
+
+## Loop Selection Flow
+
+```text
+Start
+  |
+  +-- Need strict CI/PR control? -- yes --> continuous-pr
+  |
+  +-- Need RFC decomposition? -- yes --> rfc-dag
+  |
+  +-- Need exploratory parallel generation? -- yes --> infinite
+  |
+  +-- default --> sequential
+```
+
+## Combined Pattern
+
+Recommended production stack:
+1. RFC decomposition (`ralphinho-rfc-pipeline`)
+2. quality gates (`plankton-code-quality` + `/quality-gate`)
+3. eval loop (`eval-harness`)
+4. session persistence (`nanoclaw-repl`)
+
+## Failure Modes
+
+- loop churn without measurable progress
+- repeated retries with same root cause
+- merge queue stalls
+- cost drift from unbounded escalation
+
+## Recovery
+
+- freeze loop
+- run `/harness-audit`
+- reduce scope to failing unit
+- replay with explicit acceptance criteria

package/resources/ecc/skills/continuous-learning/SKILL.md

@@ -0,0 +1,131 @@
+---
+name: continuous-learning
+description: "[DEPRECATED - use continuous-learning-v2] Legacy v1 stop-hook skill extractor. v2 is a strict superset with instinct-based, project-scoped, hook-reliable learning. Do not invoke v1; route continuous learning, session learning, and pattern extraction requests to continuous-learning-v2."
+origin: ECC
+---
+
+# Continuous Learning Skill - DEPRECATED
+
+> **DEPRECATED 2026-04-28.** Use `continuous-learning-v2` instead. v2 is a strict superset: stop-hook observation becomes PreToolUse/PostToolUse observation, full skills become atomic instincts with confidence scoring, and global-only storage becomes project-scoped plus global promotion.
+>
+> This file is kept for archival reference and backward compatibility with existing installs.
+
+---
+
+## Original v1 Documentation (archival)
+
+Automatically evaluates Claude Code sessions on end to extract reusable patterns that can be saved as learned skills.
+
+## When to Activate
+
+- Setting up automatic pattern extraction from Claude Code sessions
+- Configuring the Stop hook for session evaluation
+- Reviewing or curating learned skills in `~/.claude/skills/learned/`
+- Adjusting extraction thresholds or pattern categories
+- Comparing v1 (this) vs v2 (instinct-based) approaches
+
+## Status
+
+This v1 skill is still supported, but `continuous-learning-v2` is the preferred path for new installs. Keep v1 when you explicitly want the simpler Stop-hook extraction flow or need compatibility with older learned-skill workflows.
+
+## How It Works
+
+This skill runs as a **Stop hook** at the end of each session:
+
+1. **Session Evaluation**: Checks if session has enough messages (default: 10+)
+2. **Pattern Detection**: Identifies extractable patterns from the session
+3. **Skill Extraction**: Saves useful patterns to `~/.claude/skills/learned/`
+
+## Configuration
+
+Edit `config.json` to customize:
+
+```json
+{
+  "min_session_length": 10,
+  "extraction_threshold": "medium",
+  "auto_approve": false,
+  "learned_skills_path": "~/.claude/skills/learned/",
+  "patterns_to_detect": [
+    "error_resolution",
+    "user_corrections",
+    "workarounds",
+    "debugging_techniques",
+    "project_specific"
+  ],
+  "ignore_patterns": [
+    "simple_typos",
+    "one_time_fixes",
+    "external_api_issues"
+  ]
+}
+```
+
+## Pattern Types
+
+| Pattern | Description |
+|---------|-------------|
+| `error_resolution` | How specific errors were resolved |
+| `user_corrections` | Patterns from user corrections |
+| `workarounds` | Solutions to framework/library quirks |
+| `debugging_techniques` | Effective debugging approaches |
+| `project_specific` | Project-specific conventions |
+
+## Hook Setup
+
+Add to your `~/.claude/settings.json`:
+
+```json
+{
+  "hooks": {
+    "Stop": [{
+      "matcher": "*",
+      "hooks": [{
+        "type": "command",
+        "command": "~/.claude/skills/continuous-learning/evaluate-session.sh"
+      }]
+    }]
+  }
+}
+```
+
+## Why Stop Hook?
+
+- **Lightweight**: Runs once at session end
+- **Non-blocking**: Doesn't add latency to every message
+- **Complete context**: Has access to full session transcript
+
+## Related
+
+- [The Longform Guide](https://x.com/affaanmustafa/status/2014040193557471352) - Section on continuous learning
+- `/learn` command - Manual pattern extraction mid-session
+
+---
+
+## Comparison Notes (Research: Jan 2025)
+
+### vs Homunculus
+
+Homunculus v2 takes a more sophisticated approach:
+
+| Feature | Our Approach | Homunculus v2 |
+|---------|--------------|---------------|
+| Observation | Stop hook (end of session) | PreToolUse/PostToolUse hooks (100% reliable) |
+| Analysis | Main context | Background agent (Haiku) |
+| Granularity | Full skills | Atomic "instincts" |
+| Confidence | None | 0.3-0.9 weighted |
+| Evolution | Direct to skill | Instincts → cluster → skill/command/agent |
+| Sharing | None | Export/import instincts |
+
+**Key insight from homunculus:**
+> "v1 relied on skills to observe. Skills are probabilistic—they fire ~50-80% of the time. v2 uses hooks for observation (100% reliable) and instincts as the atomic unit of learned behavior."
+
+### Potential v2 Enhancements
+
+1. **Instinct-based learning** - Smaller, atomic behaviors with confidence scoring
+2. **Background observer** - Haiku agent analyzing in parallel
+3. **Confidence decay** - Instincts lose confidence if contradicted
+4. **Domain tagging** - code-style, testing, git, debugging, etc.
+5. **Evolution path** - Cluster related instincts into skills/commands
+
+See: `docs/continuous-learning-v2-spec.md` for full spec.

package/resources/ecc/skills/continuous-learning/config.json

@@ -0,0 +1,18 @@
+{
+  "min_session_length": 10,
+  "extraction_threshold": "medium",
+  "auto_approve": false,
+  "learned_skills_path": "~/.claude/skills/learned/",
+  "patterns_to_detect": [
+    "error_resolution",
+    "user_corrections",
+    "workarounds",
+    "debugging_techniques",
+    "project_specific"
+  ],
+  "ignore_patterns": [
+    "simple_typos",
+    "one_time_fixes",
+    "external_api_issues"
+  ]
+}

package/resources/ecc/skills/continuous-learning/evaluate-session.sh

@@ -0,0 +1,69 @@
+#!/bin/bash
+# Continuous Learning - Session Evaluator
+# Runs on Stop hook to extract reusable patterns from Claude Code sessions
+#
+# Why Stop hook instead of UserPromptSubmit:
+# - Stop runs once at session end (lightweight)
+# - UserPromptSubmit runs every message (heavy, adds latency)
+#
+# Hook config (in ~/.claude/settings.json):
+# {
+#   "hooks": {
+#     "Stop": [{
+#       "matcher": "*",
+#       "hooks": [{
+#         "type": "command",
+#         "command": "~/.claude/skills/continuous-learning/evaluate-session.sh"
+#       }]
+#     }]
+#   }
+# }
+#
+# Patterns to detect: error_resolution, debugging_techniques, workarounds, project_specific
+# Patterns to ignore: simple_typos, one_time_fixes, external_api_issues
+# Extracted skills saved to: ~/.claude/skills/learned/
+
+set -e
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+CONFIG_FILE="$SCRIPT_DIR/config.json"
+LEARNED_SKILLS_PATH="${HOME}/.claude/skills/learned"
+MIN_SESSION_LENGTH=10
+
+# Load config if exists
+if [ -f "$CONFIG_FILE" ]; then
+  if ! command -v jq &>/dev/null; then
+    echo "[ContinuousLearning] jq is required to parse config.json but not installed, using defaults" >&2
+  else
+    MIN_SESSION_LENGTH=$(jq -r '.min_session_length // 10' "$CONFIG_FILE")
+    LEARNED_SKILLS_PATH=$(jq -r '.learned_skills_path // "~/.claude/skills/learned/"' "$CONFIG_FILE" | sed "s|~|$HOME|")
+  fi
+fi
+
+# Ensure learned skills directory exists
+mkdir -p "$LEARNED_SKILLS_PATH"
+
+# Get transcript path from stdin JSON (Claude Code hook input)
+# Falls back to env var for backwards compatibility
+stdin_data=$(cat)
+transcript_path=$(echo "$stdin_data" | grep -o '"transcript_path":"[^"]*"' | head -1 | cut -d'"' -f4)
+if [ -z "$transcript_path" ]; then
+  transcript_path="${CLAUDE_TRANSCRIPT_PATH:-}"
+fi
+
+if [ -z "$transcript_path" ] || [ ! -f "$transcript_path" ]; then
+  exit 0
+fi
+
+# Count messages in session
+message_count=$(grep -c '"type":"user"' "$transcript_path" 2>/dev/null || echo "0")
+
+# Skip short sessions
+if [ "$message_count" -lt "$MIN_SESSION_LENGTH" ]; then
+  echo "[ContinuousLearning] Session too short ($message_count messages), skipping" >&2
+  exit 0
+fi
+
+# Signal to Claude that session should be evaluated for extractable patterns
+echo "[ContinuousLearning] Session has $message_count messages - evaluate for extractable patterns" >&2
+echo "[ContinuousLearning] Save learned skills to: $LEARNED_SKILLS_PATH" >&2